The Cowork Colleague Setup

The Scenario

You're a senior marketing manager who's been using Cowork for three months. It works, but every session starts from scratch. You re-explain your role, your company's brand guidelines, your formatting preferences, and the project context. Last week you asked it to draft a client email and it used American spelling, addressed the client by their surname (your company uses first names), and formatted the email in a style that looked nothing like your team's standard template.

None of this is Cowork's fault. You haven't configured it to know any of this. You've been treating a long-term colleague like a temp worker who arrives each morning with no briefing.

Today, you're going to set up Cowork properly. By the end, it'll know who you are, how your company communicates, what formats you prefer, and what standards you expect — and it'll apply all of this automatically without you repeating yourself.

Prerequisites

• Claude Desktop with Cowork enabled (Pro or Max plan)
• A clear understanding of your role, your team's communication standards, and your most common Cowork tasks
• Sample documents that represent your company's standards (email templates, report formats, brand guidelines)
• 30-60 minutes to think carefully about what you want Cowork to know about you

Step 1: Audit Your Repeat Instructions

Before creating any configuration, spend 10 minutes reviewing your recent Cowork sessions. Look for patterns:

• What do you find yourself explaining at the start of every session?
• What corrections do you make repeatedly? (Spelling, tone, format, naming conventions)
• What context do you always provide? (Team names, project names, client names, deadlines)
• What preferences do you enforce? (British English, specific heading styles, maximum document lengths)

Write a list. Common items include:

• "I always have to tell it to use British English"
• "It formats reports wrong — we use numbered sections, not bullets"
• "It never gets the client's preferred title right"
• "I keep pasting in our brand voice guidelines"
• "It uses the wrong model for simple tasks and burns through my rate limit"

This list becomes the specification for your configuration. Everything on it is a problem that context files, memory, or global instructions will solve.

Checkpoint: You have a written list of at least 8-10 repeat instructions or corrections from your recent Cowork sessions.

Step 2: Create the Project

In Claude Desktop, create a new Project. Name it something functional: "Marketing — [Your Company Name]" or "Client Communications Hub." Avoid generic names like "Work" or "Main" — if you later create projects for other workstreams, you'll need the name to mean something.

Note the project settings available:

• Project Instructions (persistent system prompt)
• Knowledge Base (uploaded documents)
• Memory (learned facts and preferences)
• Model selection

You're going to configure all of these.

Checkpoint: A new project exists with a clear, descriptive name.

Step 3: Write the About-Me Context File

Create a file called about-me.md that tells Cowork who you are and how you work. This isn't a CV — it's an operating manual for working with you.

Structure

Role and responsibilities:

• Your job title, team, and reporting line
• The 3-5 things you spend most of your time on
• Who your key stakeholders are (by name and role)

Communication style:

• How formal are your emails? Your reports? Your Slack messages?
• Do you prefer bullet points or paragraphs?
• How do you address people? (First names? Titles?)
• What tone do you use with clients versus internal colleagues?

Working preferences:

• British or American English?
• Date format (DD/MM/YYYY or MM/DD/YYYY)?
• Currency format (£ with commas, or $ with periods)?
• Document naming conventions?
• Preferred file formats?

Things Cowork should never do:

• Never use emojis in client communications
• Never assume a deadline — always ask
• Never CC people without being told to
• Never make up statistics — flag when data is needed

Be ruthlessly specific. "Professional tone" is useless. "Formal but warm — use first names, avoid jargon, one contraction per paragraph maximum, always open with context before the ask" is actionable.

Upload this file to your project's Knowledge Base.

Checkpoint: about-me.md is written with specific, actionable details and uploaded to the project Knowledge Base.

Step 4: Write the Brand Voice Context File

Create a brand-voice.md that captures how your company communicates. If you completed the Humaniser and Brand Voice Skill tutorial (Domain 3), you can reuse that file. If not, create one now:

Voice identity: A 2-3 sentence description of your company's personality in writing.

Rules:

• Approved and banned vocabulary (with alternatives)
• Sentence length guidelines
• Formatting standards (heading styles, paragraph structure)
• Tone adjustments by context (client-facing vs internal)

Examples: Include 2-3 before/after pairs showing generic writing transformed into your brand's voice.

Upload this to the Knowledge Base alongside about-me.md.

Checkpoint: brand-voice.md is written and uploaded to the Knowledge Base.

Step 5: Write the Project Instructions

Project Instructions are the persistent system prompt that shapes every conversation in this project. Unlike the Knowledge Base files (which provide reference material), Project Instructions give Cowork its behavioural directives.

Navigate to your project settings and write instructions that cover:

Identity:

You are assisting [your name], [your title] at [company name]. You're working within the marketing function. All communications you produce will represent the company's brand.

Standing rules:

• Always use British English spelling (colour, organisation, summarise)
• Format dates as DD Month YYYY (e.g., 4 April 2026)
• Use GBP (£) for all currency references
• Address all contacts by first name unless instructed otherwise
• When creating documents, use the heading hierarchy: H1 for title, H2 for sections, H3 for subsections
• Maximum 3 sentences per paragraph in emails; maximum 5 in reports

Quality gates:

• Before sending any draft, flag any statistics or claims that you can't verify from the knowledge base
• If asked to create a client-facing document, always include a draft subject line and suggest a send time
• Never fabricate testimonials, case studies, or data points

Reference the Knowledge Base:

• Refer to about-me.md for role context, stakeholder names, and communication preferences
• Refer to brand-voice.md for all writing style decisions

Checkpoint: Project Instructions are written and saved, covering identity, standing rules, quality gates, and knowledge base references.

Step 6: Configure Memory

Memory captures learned facts that persist across conversations. You can add memory entries manually or let Cowork learn them over time. For your initial setup, manually add key facts:

Navigate to Memory settings and add entries like:

• "The team consists of: [names and roles]"
• "Our biggest client is [name]. Primary contact is [name], [title]"
• "Quarterly reports are due on the 15th of the month following quarter-end"
• "The company's financial year runs April to March"
• "Board meetings are the last Thursday of each month"
• "The CEO prefers one-page summaries with an appendix for detail"

These are facts that don't change often but are referenced constantly. Storing them in memory means Cowork draws on them automatically without you needing to include them in every prompt.

Checkpoint: At least 8-10 memory entries covering team composition, key contacts, recurring deadlines, and organisational facts.

Step 7: Optimise Model Selection

Different tasks benefit from different models. Configure your project's model selection:

• Quick tasks (reformatting, simple edits, meeting note cleanup): Use the fastest available model to conserve rate limit
• Complex analysis (report writing, strategic documents, multi-source synthesis): Use the most capable model
• Creative work (brand copy, thought leadership, social media): Use the most capable model with higher temperature

If your plan allows model switching, document your model selection logic in a note within the project. This helps you maintain consistency and explains your reasoning to anyone else who uses the project.

Consider rate limit implications: running every task on the most powerful model will drain your allocation faster. Save it for work that genuinely needs it.

Checkpoint: You've documented a model selection strategy for different task types, and configured the default model for this project.

Step 8: Test the Configuration

Run three test tasks that exercise different parts of your configuration:

Test 1 — Email draft:

Draft an email to [stakeholder name from memory] about the Q2 marketing campaign timeline. We're on track but need their budget approval by next Friday.

Check: Does it use the correct name? British English? First name address? Brand voice? Correct email format?

Test 2 — Report section:

Write the executive summary for the April monthly marketing report. We spent £42,000 against a £45,000 budget, ran three campaigns, and generated 1,200 qualified leads (target was 1,000).

Check: Does it use the correct heading hierarchy? GBP formatting? Short paragraphs? The brand tone?

Test 3 — Meeting preparation:

Prepare a briefing document for Thursday's board meeting. I need to present the marketing team's Q1 results. Pull relevant context from what you know about our team, stakeholders, and reporting standards.

Check: Does it reference memory entries (board meeting timing, CEO's preference for one-page summaries)? Does it follow the formatting rules in Project Instructions?

For each test, rate the output on:

• Did it follow the Project Instructions? (Yes/Partially/No)
• Did it reference the Knowledge Base? (Yes/No)
• Did it use Memory entries? (Yes/No)
• How much correction was needed? (None / Minor edits / Significant rework)

Create a configuration effectiveness scorecard:

Interpreting Results

All three tests pass with minimal corrections: Your configuration is working. Save this scorecard as your baseline — when you update context files in future, re-run these same tests to verify nothing regressed.

One test fails consistently: The failing test type likely has a gap in your configuration. If emails fail on tone, your brand-voice.md needs email-specific rules. If reports fail on formatting, your Project Instructions need more explicit heading and paragraph rules. If the board briefing doesn't reference memory, verify the memory entries were saved correctly.

All tests produce mediocre results: Your context files are probably too vague. Go back to Steps 3-5 and make every rule more specific. Replace "professional tone" with exact sentence length limits, vocabulary rules, and before/after examples. The more specific your inputs, the more precise the outputs.

Checkpoint: Three test tasks completed. Configuration effectiveness scored for each.

Why This Matters

The configuration you've built isn't just a convenience — it's a compounding asset. Every conversation in this project benefits from the context you've set up. Over weeks and months, as memory accumulates additional learned facts and your context files get refined based on real usage, the project becomes increasingly valuable. The 50 minutes you invested today saves 5-10 minutes on every future task, accumulating into hours of reclaimed time per month.

More importantly, configured projects produce consistent output. When five people on your team each use a configured project with the same brand-voice.md and team-context.md, their AI-assisted output is consistent with each other — same tone, same format, same terminology. Without configuration, each person's output reflects Claude's generic defaults, which vary from session to session.

Expected Output

Your deliverable is a fully configured Cowork project:

• about-me.md — role context, communication style, working preferences
• brand-voice.md — voice identity, rules, vocabulary, examples
• Project Instructions — identity, standing rules, quality gates
• 8-10 Memory entries — team, clients, deadlines, organisational facts
• Model selection strategy documented
• Configuration effectiveness scorecard from three test tasks

The difference between an unconfigured Cowork session and a properly configured project is the difference between dictating to a stranger and delegating to a colleague who's worked with you for years.

Extension Challenges

1. Multiple projects — Create a second project for a different workstream (e.g., personal writing, a side project, a different department). Compare how the same prompt produces different output in each project due to different context files and instructions.

2. Onboarding a teammate — Duplicate your project configuration and modify it for a colleague with a different role. Document which elements are shared (brand voice, company facts) and which are personal (about-me, communication style).

3. Context file maintenance schedule — Set a quarterly reminder to review and update your context files. Companies change — new team members, new clients, updated brand guidelines. Document which files are most likely to go stale and create a review checklist.